Infer name mapping from expected schema

[rdsr]

For Avro, we can infer name mapping in the following scenario. As per @rdblue

1. If there is a name mapping in table metadata, pass it in
2. Otherwise, if the incoming schema has no field IDs, infer a mapping from the current schema and use it
3. Otherwise the schema has at least one column ID, so no inferred mapping should be used

[rdsr]

@rdblue I was looking at how ParquetReader chooses to assign fallback ids . I think it assigns fallback ids if at least one column does not have id. The behavior is described here org.apache.iceberg.parquet.ParquetSchemaUtil#hasIds()

With point 3 above, your approach is slightly different where if at least one column has an id we do not infer the table mapping.

Should we maintain the same behavior here?

Also, could we maintain an invariant where a file schema either has complete file ids or does not completely, it cannot have ids assigned to only some columns. If we agree to this then we can probably duplicate what Parquet does